考虑竞争环境的联邦学习多维契约激励机制研究  

Incentive mechanism for federated learning in competitive environment:multidimensional contract approach

作　　者：杨扬 殷红建 王超 Yang Yang;Yin Hongjian;Wang Chao(School of Management Engineering&Business,Hebei University of Engineering,Handan Hebei 056038,China;School of Information&Electrical Engineering,Hebei University of Engineering,Handan Hebei 056038,China)

机构地区：[1]河北工程大学管理工程与商学院,河北邯郸056038 [2]河北工程大学信息与电气工程学院,河北邯郸056038

出　　处：《计算机应用研究》2023年第10期3007-3015,共9页Application Research of Computers

基　　金：河北省高等学校人文社会科学研究项目(SQ2022085);河北省自然科学基金面上项目(A2020402013)。

摘　　要：针对现有激励机制无法满足竞争环境下联邦学习(FL)落地实施需要的现实问题,提出了一种适于竞争环境的多维契约激励机制。首先,引入了竞争强度描述数据拥有者和任务发布者之间的竞争关系;其次,创新性地设计了一种金钱和FL模型使用权相组合的MM(monetary-the FL model)组合激励;再次,在MM组合激励的基础上,分别在完全信息场景和不完全信息场景下,以任务发布者利润最大化为目标,构建了适于竞争环境的联邦学习多维契约激励机制设计优化问题;进一步,通过理论推导分析了契约可行性及契约最优性,并在此基础上给出了适于竞争环境的联邦学习最优契约设计算法;最后,实验结果表明,以MM组合激励为基础的多维契约激励机制提高了竞争环境下数据拥有者参与联邦学习的意愿。Aiming at the existing incentive mechanisms cannot satisfy the demand of implementation for federated learning(FL)in competitive environment,this paper proposed a multi-dimensional contract-based incentive mechanism for FL in competitive environment.Firstly,this paper introduced the competition intensity to measure the competition relationship between the data owners and the task publisher.Secondly,this paper designed a novel incentive form named as MM(monetary-the FL model)combined incentive,which combined monetary with usufruct of FL model.Thirdly,based on MM combined incentive,this paper constructed optimization problems of the multi-dimensional contract-based incentive mechanism design for FL in competitive environment for the task publisher’s profit maximization under complete information scenario and incomplete information scenario respectively.Finally,experimental results show that the multi-dimensional contract-based incentive mechanism based on MM combined incentive improves the willingness of data owners to participate in FL in competitive environment.

关 键 词：竞争环境 联邦学习 契约理论 激励机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]